Special Reviews

DNA Methylation and Epigenetic Inheritance in Plants and Filamentous Fungi

See allHide authors and affiliations

Science  10 Aug 2001:
Vol. 293, Issue 5532, pp. 1070-1074
DOI: 10.1126/science.293.5532.1070

This article has a correction. Please see:


Plants and filamentous fungi share with mammals enzymes responsible for DNA methylation. In these organisms, DNA methylation is associated with gene silencing and transposon control. However, plants and fungi differ from mammals in the genomic distribution, sequence specificity, and heritability of methylation. We consider the role that transposons play in establishing methylation patterns and the epigenetic consequences of their perturbation.

Epigenetic changes, so-called “epimutations,” occur because nucleotide sequence is not the only form of genetic information in the cell: Chromosomal proteins and DNA methylation can also be inherited, with important phenotypic consequences (1). In plants and filamentous fungi, genomic methylation is restricted mostly to transposons and other repeats (2, 3). In mammals, by contrast, coding sequences are methylated as well, except for the so-called CpG islands that often encompass the first exons of genes (4). This difference likely reflects the colonization of mammalian introns by transposons, and the possibility of methylation spreading into flanking exons (3). Of course, the human genome contains far less exonic DNA (<2%) than transposons (>45%) (5), which thus contribute more to the level of cytosine methylation overall (4). Naı̈vely, when transposons lose methylation they become activated, while when genes gain extra methylation, they become silenced.

Some of the first epimutations were observed by B. McClintock, who noted that transposons underwent cycles of inactivity in maize. These “changes in phase” are associated with changes in DNA methylation and are distinct from “changes in state,” which are usually sequence rearrangements (6). Both defective and intact transposons can also be modified epigenetically (“preset”) after exposure to an active transposon (6). These changes can be observed because transposon activity can influence neighboring pigment genes, in some cases through an outward reading promoter (7, 8). The influence of transposons on nearby genes has been documented in Neurospora crassa (9) and occurs in mouse, bacteria, yeast andDrosophila (10–13). Thus, changes in phase could have genome-wide consequences when defective elements reach high copy numbers (Fig. 1) (14). Epimutations in maize genes were first studied by R. A. Brink, and often result from allele interactions, known as paramutation (15). Genetically, paramutation resembles presetting, in that one allele (paramutagenic) can lead to a heritable change of expression of another allele (paramutable) of the same gene in heterozygotes. The presence of transposons near paramutable genes has revived the proposition that paramutation may be caused by transposons (16–19), though their role, if any, has yet to be established.

Figure 1

Transposons in maize can co-ordinate the regulation of genes. Leaf samples from plants in which loss of transposon activity results in variegation. Unlinked mutations responsible for white lesions (left) or pale green pigmentation (right) are suppressed together in double mutant plants (middle). This is because both genes have been brought under the control of the same family of transposable elements (14).

Silent transgenes were among the first examples of epimutations to be observed in plants other than maize (20), but more recently two examples of endogenous epimutations have been examined inArabidopsis. Epialleles of the zinc-finger floral geneSUPERMAN (known as clark kent or clkalleles) are densely methylated in the transcribed portion of the gene and in a 65-bp hairpin-forming CpT microsatellite surrounding the transcriptional start site (21). Likewise, epialleles of thePAI2 gene, which encodes an enzyme involved in tryptophan biosythesis, are densely methylated and inactive, but only in strains containing an inverted and methylated duplication of the closely related gene PAI1. As with SUP, methylation atPAI2 extends upstream of the transcriptional start site and ends in the 3′ untranslated region (22, 23).

In fungi, transgene duplications are efficiently methylated by RIP (repeat induced point mutation) in Neurospora crassa(24) and MIP (methylation induced premeiotically) inAscobolus immersus (25). De novo methylation is triggered in haploid nuclei, most likely by DNA pairing between repeated sequences, and is maintained vegetatively after the duplicated sequences segregate. Methylated genes behave as epimutant alleles (26, 27), although RIP leads principally to genetic mutations (24). Unlike in plants and animals, methylation-associated silencing has been shown to block transcript elongation in these fungi (3, 28,29). In Neurospora, short TpA segments can act as portable signals for methylation (30), while inAscobolus, transfer of DNA methylation between alleles is related to homologous recombination (31). As yet there is no evidence for either process in plants, but the implications for paramutation and transposon silencing are clear. Most transposons in fungi are heavily methylated and inactivated, with the only exceptions being too short to be targeted by MIP or RIP (24, 32).

DNA Methyltransferases and Methylation Mutants

Cytosine methyltransferase enzymes catalyze the transfer of an activated methyl group from S-adenosyl methionine to the 5 position of the cytosine ring (5-Me-C). In mammals, two distinct types of methyltransferases, Dnmt1 and Dnmt3, have been functionally characterized (33). Dnmt1 is thought to be primarily involved in maintenance and Dnmt3 in establishment of genomic methylation patterns (33). In Ascobolus, themasc2 gene was identified by its similarily toDnmt1. The masc2 gene product has methyltransferase activity in vitro, but a knockout masc2 mutation displayed no obvious effect on MIP methylation or maintenance of genomic methylation patterns (34). Conversely, another predictedAscobolus methyltransferase, masc1, did not display any in vitro activity, but knockout mutants were defective for MIP methylation (35). However, neither the single masc1 mutant nor the masc1 masc2 double mutant was defective for maintenance of genomic methylation patterns, suggesting that otherAscobolus methyltransferases remain to be identified (34, 35). In Neurospora, the dim-2 methyltransferase was identified by positional cloning of a methylation-deficient dim-2 mutation (36). The dim-2 mutant can still perform the point mutations associated with RIP however (36). A second Neurospora gene, related to masc1, has been found by genome sequencing, and this gene product might control RIP (36).

Arabidopsis has at least 10 genes that could encode DNA methyltransferases, more than any other eukaryote sequenced so far (37). The MET class encodes genes related to the mammalianDnmt1 (38). MET1 has been functionally characterized in a transgenic line where its expression is suppressed by an antisense-MET1 transgene. Antisense-MET1plants have as little as 10% of the wild-type levels of 5-Me-C (39, 40). Mutants of the MET1 gene (ddm2 mutants) have also been identified by genetic screens for plants with reduced methylation of repeated sequences (41, 42). Three MET1-related genes,MET2a, MET2b, and MET3, remain to be functionally characterized (38). A second type of methyltransferase, the CMT “chromomethylase” class, is also related to Dnmt1 except that a novel chromo-domain amino acid motif is inserted between two of the canonical methyltransferase motifs (43). This class is unique to plants, and there are three related CMT genes in Arabidopsis (38).CMT1 is predicted to be nonfunctional in all strains characterized (43), but CMT3 has been shown to be functional (44–46). A third class ofArabidopsis methyltransferase is the “domain rearranged methyltransferase” or DRM class, which is most related to Dnmt3, except that the canonical methyltransferase motifs are organized in a novel order (47). Two predicted DRM genes,DRM1 and DRM2, have not yet been characterized for function. However, it is attractive to think that, like the mammalian Dnmt3 enzymes, they might be involved in the establishment of methylation patterns. Finally, one gene (GenBank accession number At5g25480) resembles Dnmt2, a highly conserved but enigmatic putative mammalian methyltransferase gene, with homologs inSchizosaccharomyces pombe and Drosophila, and no known function (48).

In Arabidopsis, screens for mutants with demethylated DNA (decrease in DNA methylation or ddm mutants) were conducted in a similar manner to those in Neurospora, and multiple mutations were recovered at two loci (41).DDM1 encodes a likely SNF2/SWI2 chromatin remodeling protein (49), while DDM2 encodes MET1(42). Interestingly, when demethylated sequences fromddm1 or antisense-MET1 backgrounds are inherited independent from the methylation defect, they remain demethylated. This observation suggests that de novo methyltransferase activity in plants is quite low, and that mutants might have little phenotype (41). In mouse embryos, by contrast, genome demethylation and remethylation occur over a few cleavage divisions, and mutations in the de novo methylase DNMT3 are responsible for the ICF syndrome in humans (immunodeficiency, centromere instability, and facial anomalies) (50, 51).

Epigenetic Consequences of DNA Methylation Mutants

In Neurospora, the dim-2 mutation has no obvious phenotype under laboratory conditions despite genome-wide demethylation (36). In Ascobolus, themasc1 mutation also has no obvious phenotype during vegetative growth, but it blocks MIP methylation and confers sterility in masc1 homozygous dikaryons (35). The effects of methylation defects are more conspicuous in plants. For example, in antisense-MET1 strains of Arabidopsis, flowering time, fertility, leaf and floral morphology are all heritably altered (39, 40, 52). ddm1mutants display similar phenotypes, but often take many generations of inbreeding to manifest these changes (53). The molecular defects induced by both ddm1 and antisense-MET1are varied. They include demethylation and ectopic expression of methylated genes, demethylation and movement of transposons, and local hypermethylation and silencing of genes.

Demethylation of a tandem repeat in the promoter accompanies the ectopic expression of the late flowering gene fwa inddm1 plants, and this results in a dominant epimutation (56). Similarly, ddm1 targets a complex pathogen resistance gene cluster containing heavily methylated retrotransposons, leading to elevated expression in the dwarf epimutantbal (42). In a third example, a new allele ofdwarf4 arose in a single ddm1 line. This mutation (clam) was caused by insertion of a CACTA transposable element, related to maize Suppressor-Mutator(55). In fact, several classes of transposons are demethylated and activated in ddm1. While both CACTA and MULE (Mutator-like) DNA transposons are mobilized, leading to a spectrum of new insertions (55, 56), heterochromatic retrotransposons related to Athila andTar17 do not appear to transpose in ddm1 strains, despite being transcriptionally activated (57).

These seemingly disparate cases may be related by the nature of heterochromatin. Long arrays of tandem repeats may be responsible for heterochromatic gene silencing (58). By analogy,fwa and bal are associated with methylated repeats and may represent what B. McClintock called “cryptic” heterochromatin (59). Consistent with this view, pericentromeric and knob repeats, including many transposons, are also demethylated in ddm1 (43), and this has led to the suggestion that DDM1 acts to enable access of DNA methyltransferases to heterochromatin (49). Thus, activation of both conspicuous and cryptic heterochromatin in ddm1mutants could lead to the variety of defects found.

Transgene silencing was relieved in both ddm1 andddm2 (or met1) mutants, although only those transgenes that were silenced at the level of transcript initiation were uniformly activated in ddm1 (61, 62). Screens for mutants that relieved this type of silencing recovered at least 12 mutations in 5 genes, including DDM1 (57, 61,63). hog1 mutants resulted in a significant reduction in DNA methylation (63), but mom1,sil1 and sil2 mutants did not (57,63). These may encode chromatin modifying enzymes, a suggestion supported by cloning of the SNF2-related geneMOM1 (57). DNA methylation mutants also affect endogenous epimutations. For example, PAI2 silencing and methylation is relieved in ddm1 (64), but only in strains without the PAI1-PAI4 inverted duplication (65). Methylation of PAI1-PAI4 was more reduced in ddm2, but this had less effect on methylation atPAI2 (65).

Asymmetric Methylation and Antisense RNA

While methylation in mammalian genomes is generally restricted to CpG sequences, plant and fungal genomes have methylated cytosines in other contexts as well. The SUP and PAI epialleles, for example, have extensive asymmetric and CpXpG methylation (21, 23). Two genetic screens have been used to address the origin of this methylation and both have identified the chromomethylase gene CMT3 (44,45). Analysis of methylation patterns using bisulphite sequencing show that cmt3 primarily affects CpXpG methylation and, furthermore, CMT3 transgenes restore methylation in cmt3 plants (45), apparently de novo, although some methylation is left intact and may be used as a mark for remethylation. This residual methylation must be generated by yet another methylase, and CMT2,Dnmt2-, and Dnmt3-like genes are good candidates. Intriguingly, antisense CMT2 transgenes failed to inactivate the endogenous gene in over 50 transgenic plants (66), consistent with a requirement forCMT2 in antisense silencing. Chromodomains inDrosophila direct proteins to heterochromatin and have been shown in one case to bind untranslated RNA (67). This has led to the speculation that chromomethylases might mediate RNA-directed DNA methylation (68, 69), which has been demonstrated to be predominantly asymmetric in plants (70, 71).

Thus, while DNA-DNA pairing is certainly possible (21,72), antisense transcription of SUPand PAI might be responsible for the asymmetric methylation. In the case of PAI1-PAI4, the inverted duplication is tail-to-tail, and would be expected to generate a hairpin transcript (73). In the case of SUP, promoter methylation might be triggered by antisense transcription of the hairpin-forming CpT microsatellite at the start site of transcription, in either the endogenous or transgenic copies of the gene. Inverted repeats have been shown to be good targets for DNA methylation in Arabidopsis, and perhaps this is why (23). For similar reasons, RNA-mediated methylation is also an attractive model for paramutation and presetting in maize.

Paradoxically, methylation actually increases at SUP andAG in met1 and ddm1 mutants, leading to epimutant alleles at a very high frequency (52). Could this also be due to RNA-mediated methylation? Perhaps transposons such as MITES (miniature invert repeat transposable elements), which reside in the 3′ untranslated regions of many plant genes, initiate antisense transcription when they are activated in methylation mutants. Alternatively, short transcripts produced from CpT repeats elsewhere in the genome might target methylation at these genes. Both AG and SUP have repeats related to simple sequences found in other genes (52).

RNA-dependent DNA methyltransferases may serve to monitor transcribed transposable elements. Transposons generate dsRNA because of their random integration in the genome, and because they have promoters in their terminal repeats (74). For this reason, they appear to be major targets of RNA interference inCaenorhabditis elegans (75,76). In plants, they may be targeted for methylation using RNA as a guide. This methylation would be maintained over generations, resulting in the differential methylation of transposons and genes. Asymmetric methylation has been found during early embryogenesis in mammals (77) when the components of RNA interference are also active (78). Perhaps mammalian exons are methylated by transcriptions from transposons in neighbouring introns. However, chromomethylases are not found in mammals or fungi, indicating some other methyltransferase must be responsible for de novo transposon methylation in these genomes.

In Ascobolus and Neurospora, candidate RNA-dependent DNA methyltransferases have not been found. This could explain why quelling (an RNA-mediated process) requires the continuous presence of a silencing transgene array (which may become methylated) but does not lead to methylation of the endogenous target gene (79). Nonetheless, asymmetric methylation is found in both fungi, suggesting that it is established and maintained by different mechanisms (24, 80).

A Role for Silencing and DNA Modification in Development

Mutants in the silencing genes DDM1, MOM1,CMT3, HOG1, SIL1 and SIL2, have very minor effects on plant development as a whole. Even mutants in MET1 have relatively mild phenotypes that can be explained by generation of epialleles elsewhere in the genome. Similarly, dim-2 has no detectable phenotype inNeurospora. In mammalian cells, by contrast, differentiation of methyltransferase negative embryonic stem cells leads to rapid programmed cell death, and this may require p53 (81), which is absent from plants and fungi (37).

Although redundancy may explain the absence of strong phenotypes in plants, what can we learn about the biological role of DNA methylation, gene silencing, and chromatin modification from these mutants? It has been reported that the paternal genome is silenced throughout much of early embryonic development inArabidopsis (82). However, at least one gene used in this study is expressed from the paternal genome at the earliest detectable stage (83), indicating that transcriptional gene silencing is unlikely to be involved. Nonetheless, thepolycomb genes MEDEA/FIS1 and FIE1, which are required for seed development, are not expressed from paternal chromosomes in the endosperm. fie1 is partially rescued by anti-met1 transgenes, and there is some evidence that ddm1 ameliorates medea after fertilization (84, 85). Consistent with these observations, anti-MET1 plants appear to be more tolerant of changes in parental dosage in the developing endosperm, a phenomenon related to imprinting (86).

As in Drosophila, mutants that destabilize endogenous gene repression have been found in plants. For example, curly leaf (a polycomb homolog), pickle (aMi-2-like SNF2 homolog) and asymmetric leaves1 (a myb domain homolog) derepress homeotic gene expression and have strong phenotypic effects (87). Unlike ddm1, however, they do not affect genomewide methylation, supporting the notion that heterochromatin is the primary target of DDM1, rather than euchromatic genes.

Given that methylation is prevalent in heterochromatin, and genes affecting it have little developmental role in plants and fungi, what then is its function? An obvious candidate is transposon regulation. While it is clear that transposons can be highly detrimental, they typically accumulate in heterochromatin ensuring a minimal impact on gene expression. This is true both of Arabidopsis andNeurospora, though transposons are far more widely distributed in maize (37, 88, 89). Euchromatic insertions are observed in Arabidopsis, but they are only widespread inddm1 mutants (55, 56). This indicates that DDM1 might be required to target transposons to heterochromatin, and that it may be differently regulated, or have reduced function, in maize.

Telomeres and centromeres, which are uniquely eukaryotic chromosomal components, not only accumulate transposons but also share several properties with them, such as replication through reverse transcription in the case of telomeres and LINEs. Retroelements are the largest single component of most eukaryotic genomes and although the analogy between the nucleus and a retroviral particle is perhaps far-fetched, the notion that centromeres and telomeres were derived from transposons is gaining some appeal.

Epigenetic Inheritance and Natural Variation

The expression level of a gene—rather than the sequence of the protein product—can often determine phenotypes that contribute to natural variation (90). Epigenetic modifications provide a means of altering expression states. For example, methylation and silencing of the cycloidea gene of the plant Linaria vulgaris causes a flower morphology change. This methylation state arose approximately 250 years ago, and has been maintained by selection ever since (91). It has been proposed that epigenetic differences revealed by ddm1might contribute to the loss of fitness observed upon inbreeding (53). The fact that such differences are polymorphic between strains provides a potential explanation for hybrid vigor and suggests that transposons may be involved (92), although unlike epimutations, new transposon insertions are polymorphic both within and between strains (56). Epigenetic changes have also been implicated in the genome remodeling that occurs in allotetraploid plants, where two related but distinct diploid genomes are combined in a single nucleus. For example, in allotetraploids made between different diploid Brassicas, or between Arabidopsis and a related species Cardaminopsis, nucleolar dominance is observed. In this phenomenon, the rDNA genes in the tetraploid are expressed from only one of the two parental diploid genomes. The dominance can be reversed by the methyltransferase inhibitor compound 5-azacytidine, suggesting a role for methylation (93). Individual genes and transposon sequences can also be silenced from one of the two genomes in a tetraploid (94,95). The application of genome-wide expression profiling to polyploids in wild- type and methylation-deficient mutant backgrounds may reveal the molecular basis for these epigenetic changes and contribute to our understanding of natural variation in plants.

  • * To whom correspondence should be addressed. E-mail: martiens{at}cshl.org (R.A.M); colot{at}evry.inra.fr (V.C.).

  • Permanent address: Unité de Recherches en Génomique Végétale (INRA-CNRS), 2 rue Gaston Crémieux, CP 5708, 91057 Evry Cedex, France.


View Abstract

Navigate This Article